MCAM abnormal expression and clinical outcome associations are highly cancer dependent as revealed through pan-cancer analysis

Abstract

MCAM (CD146) is a cell surface adhesion molecule that has been reported to promote cancer development, progression and metastasis and is considered as a potential tumor biomarker and therapeutic target. However, inconsistent reports exist, and its clinical value is yet to be confirmed. Here we took advantage of several large genomic data collections (Genotype-Tissue Expression, The Cancer Genome Atlas and Cancer Cell Line Encyclopedia) and comprehensively analyzed MCAM expression in thousands of normal and cancer samples and cell lines along with their clinical phenotypes and drug response information. Our results show that MCAM is very highly expressed in large vessel tissues while majority of tissues have low or minimal expression. Its expression is dramatically increased in a few tumors but significantly decreased in most other tumors relative to their pairing normal tissues. Increased MCAM expression is associated with a higher tumor stage and worse patient survival for some less common tumors but not for major ones. Higher MCAM expression in primary tumors may be complicated by tumor-associated or normal stromal blood vessels yet its significance may differ from the one from cancer cells. MCAM expression is weakly associated with the response to a few small molecular drugs and the association with targeted anti-BRAF agents suggests its involvement in that pathway which warrants further investigation.

Introduction

MCAM, better known as CD146, is a member of the immunoglobulin superfamily and functions as a Ca2+ independent adhesion molecule [1]. Although the 1st report of its expression in melanoma and its association with tumor progression and metastasis attracted more attention [2, 3], accumulating evidence has showed that it has much broader and more complex functions with involvement of many physiological and pathological processes [4]. Not only it is primarily expressed in the intercellular junction of endothelial cells of blood vessels but also participates in the regulation of organ development of nervous system [5], kidney [6] and retina [7]; immune response; and angiogenesis [8] and lymphangiogenesis [9].

MCAM research in tumors has been very active in recent years as growing evidence points it to a potential therapeutic target. Increased MCAM expression has been reported in many solid tumors beyond melanoma, and it is often associated with tumor progression, metastasis and poor outcome of many different cancers such as breast cancer [10–12], clear cell renal cell carcinoma (RCC) [13], colorectal cancer [14], esophageal squamous cell carcinoma [15], Ewing sarcoma [16], hepatocellular carcinoma [17, 18], gastric cancer [19], gallbladder adenocarcinoma [20], leiomyosarcoma [21], lung cancer [22, 23], pancreatic cancer [24] and ovarian cancer [25, 26]. A meta-analysis from 12 studies with 2694 patients found strong significant associations between MCAM expression and both overall survival (hazard ratio of 2.5) and time to progression (2.45) [27]. Laboratory data also showed that blocking MCAM inhibited tumor growth and tumor angiogenesis in xenograft model of melanoma, pancreatic and colon cancer [28, 29], which provides sound rationale for its therapeutic value.

In spite of the advances, inconsistencies and confusions exist. A more recent study showed that reduced MCAM expression promoted, not inhibited as previously reported, tumorigenesis and cancer stemness in colorectal cancer, and this effect was likely mediated through activating Wnt/β-catenin signaling [30]. Another study specifically focused on gene expression of vascular endothelial cells and found that although RCC, colorectal carcinoma or colorectal liver metastasis all had high and consistent expression of MCAM, this higher expression was only associated with poor survival of RCC, not in colorectal carcinoma [31]. Such association was clearly from tumor-associated angiogenesis, not from cancer cells per se. As most of previous studies were conducted by immunohistobiochemistry, MCAM expression is presumably localized in cancer cells, which leads to the question whether the source of MCAM expression, from tumor cells or from endothelial cells of blood vessels, matters in cancer biology and clinical outcomes and contributes to the result discrepancy. Furthermore, most of previous studies were conducted on a small number of patients and from certain regions of the world; selection bias or geographical difference might also underlie the difference.

Taking advantage of the large data from Genotype-Tissue Expression (GTEx; https://www.gtexportal.org/home/), The Cancer Genome Atlas (TCGA; https://portal.gdc.cancer.gov/) and Cancer Cell Line Encyclopedia (CCLE; https://portals.broadinstitute.org/ccle/), we systematically analyzed the expression pattern of MCAM across 53 different normal human tissues, 32 types of different tumors along with their paired normal tissues and over 1000 cancer cell lines. The association of its expression with tumor progression and patient survival was explored for all cancers and each cancer separately. MCAM expression in pure cancer cells of CCLE was investigated to answer the question of intrinsic expression of cancer cells versus tumor-associated blood vessel endothelial cells. The expression of MCAM was also correlated to cell response to 544 drugs for its potential as a therapeutic target. This 1st-ever comprehensive analysis of MCAM showed that MCAM expression and implications were highly tissue or cancer dependent. Its expression in most major cancers was actually decreased relative to their normal tissues and did not show an increasing trend with a higher tumor stage. However, the results also point us some new directions on which tumors to focus on and some critical questions to address.

Methods and materials

MCAM expression in normal tissues

MCAM expression in normal tissues was examined in RNA-seq data from GTEx V7 release (https://www.gtexportal.org/home/), which consists of about 11600 samples from 53 tissue types/organs. The pre-processed data were downloaded from the portal, and transcripts per million (TPM) was used to compare the relative expression among different tissues/organs or sub-locations of an organ.

TCGA data for primary tumors

Processed and normalized gene expression data (upper quartile of 1000 normalized) generated by RNA-seq were downloaded from TCGA data portal (https://portal.gdc.cancer.gov/), which contains 11069 samples and 20531 genes for 32 cancer types. Clinical data were also downloaded, cleaned and formatted for the analyses. Only expression data for MCAM were extracted for in-depth analysis. For differential expression between tumors and normal samples in each cancer type, the expression data were log2 transformed and two-group t-test was performed for those tumor types with at least two normal samples. The log2 fold change and significant P-value (minus log10) were plotted by Volcano plot for each cancer type. Tumors with t-test P-value less than 0.01 were considered as differentially expressed. Of note, some tumors had very few paired normal samples and were significantly underpowered to detect differential expression. MCAM expression among different stages of cancer (TNM stage) was evaluated by ANOVA test and P-value less than 0.05 was considered as significant. For survival association analysis of MCAM, Cox proportional hazard model was used to model log2 MCAM expression and time to death where a patient alive at the last follow-up was censored. The hazard ratio and 95% confident intervals were plotted by R package ‘forestplot’. The interval without overlapping with 1 indicates significant association. The ratio greater than 1 means increased expression is associated with worse survival (less than 1 is favorable) and the numeric value represents an increased (or decreased) risk for every doubling expression of MCAM.

Cell line MCAM expression and drug response

CCLE is a project to conduct a detailed genetic characterization of a large set of human cancer cell lines for DNA mutation, copy number and mRNA expression for about 1000 cancer cell lines. We downloaded the mRNA expression data from the project data portal (https://portals.broadinstitute.org/ccle/), which contains normalized reads per kilobase million data for 1076 cell lines. The drug response data were downloaded from The Cancer Therapeutics Response Portal (CTRP, https://portals.broadinstitute.org/ctrp.v2.1/) with response profiles for 544 drugs in ~900 cell lines with RNA-seq expression data. We checked differential expression of MCAM among different types of cell lines and conducted correlation analysis between MCAM expression and drug response area under the curve (AUC) first for all cell lines together and then separately for each cancer cell line type with at least 30 cell line replicates by Pearson correlation. The P-value less than 0.05 was considered as significant. For each cancer cell line type, we also obtained the percentage of drugs that had significant association with MCAM expression, another way to delineate the potential cancer type specific benefit by targeting MCAM.

All data analyses were conducted in R Version 3.4.2 (https://www.r-project.org/) unless stated otherwise.

Results

MCAM is highly expressed in large vessels while low in most other normal tissues

To get better understanding of MCAM expression and distribution in the normal human tissues, we first examined its expression across 53 different tissues in GTEx data (https://www.gtexportal.org/home/). As shown in Figure 1, the expression of MCAM is highly variable across different tissues, with the median expression levels across tissue types ranging from 0.8 (whole blood) to 1486.4 (tibial artery) TPM. The highest expressed tissue is a large artery in which tibial, coronary and aorta are the top three tissue types. Most other tissues have relatively low and minimal expression. For the non-artery tissue, adipose and esophageal (gastroesophageal junction and muscle layer, not epithelial) tissues have relatively high expression. Brain tissues, regardless of brain locations, have very minimal expression.

MCAM differential expression between tumors and their adjacent normal tissues is highly dependent on tumor type

We profiled MCAM expression in 32 types of tumors from TCGA. We first plotted its expression distribution across all tumors and their normal tissues where available (Figure 2A). All tumors had a certain level of expression but the expression in kidney renal clear cell carcinoma (KIRC), pheochromocytoma and paraganglioma, sarcoma and skin cutaneous melanoma (SKCM) was much higher than other cancer types. MCAM was initially discovered from melanoma and was found to be associated with its progression and metastasis [2, 3]. The high expression of MCAM in this tumor in the TCGA dataset suggests mRNA expression is a reliable measurement for MCAM expression and confirms its important role in melanoma. For tumors with at least two normal tissues sequenced, we conducted differential expression analysis. Surprisingly, most tumors (14 of them) had reduced MCAM expression in the tumors compared to their normal tissues although 7 tumor types had increased expression; however, only 6 and 5 tumor types were differentially expressed statistically, respectively (Figure 2B). SKCM, the most highly expressed tumor, only had one normal tissue and statistical analysis was not possible. MCAM was most differentially expressed in KIRC with more than 4-fold increase in the tumors compared to their paired normal kidney tissues (P-value of 3.10E-37). On the reduced expression side, breast invasive carcinoma and lung squamous cell carcinoma were the top two tumors with more than 2-fold decrease. It is noted that some tumors only had a limited number of normal samples and there was not sufficient statistical power to detect a significant difference. For example, cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC) only had two paired normal samples with a differential P-value 0.02; however, the MCAM expression had nearly 8-fold decrease in tumors compared to the normal samples.

Association of MCAM expression in tumors with clinical phenotypes

MCAM was reported to be associated with melanoma metastasis; thus, we first examined the differential expression of MCAM across the different stages of this tumor. Although there was a slight increase in stage III and IV tumors versus stage I and II tumors, there was no statistical significance between any stage (ANOVA P = 0.35, Figure 3A). Further noted is that MCAM expression in stage IV tumors is indeed less variable and tends to be higher.

When MCAM expression was analyzed between stages for all tumors together, a significant difference was observed between stage I and stage II/III, stage II and III and stage II and IV but not between stage I and stage IV; however, higher stages did not correlate with increased expression and also the difference between stages was quite small (Figure 3B).

Stage association was further analyzed for each tumor separately and only six tumors whose MCAM expression was significantly associated with tumor stage, which includes adrenocortical carcinoma (ACC), bladder urothelial carcinoma (BLCA), colon adenocarcinoma, KIRC, testicular germ cell tumors and thyroid carcinoma (Figure 4A). Although MCAM expression was generally increased in these tumors’ later stages, the pattern was not consistent as in some tumors its expression in an earlier stage was seen higher than in a later stage such as in ACC.

We further conducted survival association analysis between MCAM expression and overall survival for each tumor type. This result showed that MCAM expression was not associated with overall survival in most tumors except for BLCA, CESC, brain lower grade glioma (LGG), mesothelioma (MESO), stomach adenocarcinoma (STAD) and uterine corpus endometrial carcinoma (UCEC; Figure 4B). Interestingly, skin melanoma (SKCM) had a hazard ratio of 1.04 with the tightest confidence intervals. As its lower bound of 95% confidence interval overlaps with 1, it is marginally significant. The significant association of LGG is worth attention. In normal brain tissues there was almost no MCAM expression in GTEx dataset; however, its expression in brain tumor is significantly increased. The hazard ratio of 1.5 has very tight confident intervals and represents 50% increased risk of death when MCAM increases 1-fold.

MCAM expression in tumor cell lines

One of the challenges in tissue-based RNA expression is the confounding from non-tumor cells or tissue, which can be highly variable from tumor to tumor. To resolve the issue, we sought the CCLE RNA-seq data from over 1000 cell lines derived from 26 different tumors. We first checked MCAM expression across different types of cell lines and found that the cell lines from SKCM still had the highest expression of MCAM, consistent with the finding in TCGA tissue samples. Cell lines from brain tumors also had high expression of MCAM, contrast to its expression in normal brain tissues whose expression was very low in GTEx dataset. This also suggests that the increased MCAM expression in TCGA LGG was indeed from tumor cells. Hematological cell lines like AML had very low expression (Figure 5A). The expression of MCAM appears highly correlated with promoter DNA methylation as shown from the matching reduced representation Bisulfite sequencing (RRBS) data for these cell lines. For example, melanoma cell lines had the lowest promoter methylation where its expression was the highest. Burkitt lymphoma had very low expression MCAM yet its promoter methylation was the highest (Figure 5B).

MCAM expression and drug response

We evaluated the association of MCAM expression in 815 cell lines with the response data to 545 drugs and found that only two drugs, ‘PLX4720’ and ‘dabrafenib’, had a correlation coefficient greater than 0.3 (both are negatively correlated; Figure 6A). Higher expression of this gene was correlated with better response (smaller response AUC value). Interestingly, both PLX4720 and dabrafenib are targeted therapy agents to treat melanoma with BRAF gene mutation, particularly for metastatic melanoma. Six other drugs had a correlation coefficient greater than 0.2 but only one (Fluorouracil) had positive correlation (higher expression of MCAM associated with resistance).

For cell line types with at least 30 cell lines, we further correlated MCAM expression with each of 545 drugs separately (a total of 10 cancer cell line types) and calculated their significant drug response ratios among the 545 drugs. Although MCAM expression was correlated with at least one drug in all these different cancer lines, the response rates were quite different and highly cell line type dependent (Figure 6B). While there were 10% of drugs whose responses were correlated with MCAM expression in upper aerodigestive tumor cell lines, cell lines from hematopoietic/lymphoid and lung cancer only had less than 1% drugs whose responses were correlated with MCAM expression. Lack of correlation in hematopoietic/lymphoid cell lines is likely explained by their very low expression in these cell lines although it is not the case for lung cancer cell lines as MCAM expression in these cell lines are about the similar range as aerodigestive tumor cell lines (Figure 5A).

Discussion

The implication of MCAM in tumor biology and clinical significance has attracted lots of attention. Many studies report that its overexpression is significantly associated with tumor progression, angiogenesis, metastasis or poor patient outcome, a potential biomarker or therapeutic target. However, conflicting results exist. Most previous studies have a limited sample size on a few selected tumors; thus, its clinical association is often underpowered or unstable. Moreover, its expression and clinical implication in a full spectrum of tumors are little known.

In this study, we conducted comprehensive analysis of MCAM across 53 normal tissue types, 32 tumor types and nearly 1000 cancer cell lines. Not limited to its relative expression across different tissues and tumors, we also profiled its clinical associations with reported variables such as tumor stage, patient overall survival or as a potential drug target. First, we found that MCAM is highly expressed on large vessels like tibial, coronary and aorta vessels but very low in most other normal tissues. Second, MCAM expression was increased in seven tumors relative to their paired normal tissues in TCGA collection while most others were decreased in tumor tissues. Next we also found that MCAM expression had no significant statistical association with tumor stage in SKCM although it was significant when all tumors were analyzed together. The latter was mostly contributed by dramatically increased sample sizes so that even very small difference could be detected, which may not biologically and clinically significant. It was also not in increasing trend along with increasing stage. The weak association was found in six rarely reported tumors (BLCA, CESC, LGG, MESO, STAD and UCEC). For extensively studied patient survival association, we indeed found that higher expression of MCAM was associated with poor survival of patients in melanoma although the effect was small (4% increased risk with doubling expression) and the statistical significance was marginal. Interestingly we did not observe significant association in some commonly reported cancers such as lung, kidney and liver although significant association was validated in gastric cancer as previously reported [19]. In our cancer cell line analysis, aimed to distinguish intrinsic tumor MCAM expression from stromal cells, we found consistent high expression of MCAM in melanoma cells but also noticed some other cells not commonly studied with high expression and potential significance. For example, normal brain tissues had minimal expression of MCAM in GTEx data; however, glioma and meningioma became the 2nd and 4th highly expressed tumor cells. Also interesting is that this expression was highly correlated with promoter methylation of these cells.

The result discrepancies with literature reporting could be from several sources. Most previous studies used Immunohistochemistry (IHC) to measure positivity of MCAM expression [27]. The IHC reflects the protein expression rather than mRNA expression from RNA-seq and it is possible that mRNA expression may not be correlated with protein expression. Our data from the TCGA SKCM appear refuting the possibility, as the results for melanoma are largely consistent with previous reports in terms of expression and clinical phenotype association. The RNA-seq data from this study are also highly concordant with a previous study where both IHC and Reverse transcription polymerase chain reaction (RT-PCR) were conducted for MCAM on liver cancer and both showed increased MCAM expression in cancer tissues/cells compared to normal liver cells [17]. The significant association of MCAM expression and patient overall survival was also reported in purely RNA-based assay in renal cell carcinoma [13]. Noted is that for most common cancers in TCGA the sample sizes are generally greater than 400 and the lack of association is unlikely the result of insufficient power. Unlike IHC where MCAM expression is semi-quantitative where expression level classification can be arbitrary, mRNA-seq data is continuous and more sensitive to detect genes at low expression and association.

The source of MCAM expression, either from tumor cells or from tumor-associated or stromal blood vessels, appears important to distinguish as their relevance to tumor phenotype behaviors may differ. Renal cell carcinoma was one of the tumors with very high expression of MCAM (only second to melanoma) in TCGA samples; however, its expression in cell lines was around the middle, which suggests that its high expression in primary tumors was partially contributed by non-tumor cells such as blood vessels. This was supported by a previous study specially focused on tumor-associated blood vessels [31]. Regardless of the source, that study showed that increased expression of MCAM in blood vessel was also associated with poor outcome. Melanoma is the only tumor that showed the highest MCAM expression in both primary tumors and cell lines and this expression is likely to be entirely from tumor cells. The confounding from normal tissue may lead to conflicting reports and explain its down-expression in many primary tumors such as lung cancer where normal lung has rich blood vessels. This reminds us that using increased expression in tumors relative to their adjacent normal tissues to define an oncogenic gene may not be accurate in some cases.

Although this study focused on TCGA data for MCAM expression in tumor samples, we also mined the data in Oncomine (https://www.oncomine.org/resource/login.html), the resources with a large collection of microarray-based gene expression datasets from many different research groups. In spite of the heterogeneous nature, we found many striking similarities. For example, all studies with MCAM differential expression between tumors and normal tissues for lung (seven studies), bladder (three) and prostate (six) show reduced MCAM expression in tumors while studies for esophageal (three), gastric (six), head neck (five), liver (two) and lymphoma (seven) demonstrate increased expression. Six out of seven studies in kidney were found with increased MCAM expression and the one without is due to the special type of kidney cancer ‘Chromophobe Renal Cell Carcinoma’. Survival association analysis is less consistent and often hindered by lack of information and a limited sample size. For selected cancer like lung cancer, only 1 of 10 studies shows increased MCAM expression was associated with worse patient overall survival (P = 0.041). These additional data mining suggests that the results we presented here from TCGA are convincing, and the significant role of MCAM may not be that broad but limited to a few cancers.

The significant association of MCAM expression with PLX4720 and dabrafenibin treatment response in cell line data is interesting. Both drugs target BRAF mutation in targeted therapy for melanoma and whether the drugs act through MCAM or MCAM is just a surrogate needs further investigation. There is no significant co-expression between MCAM and BRAF in both CCLE cell line data (correlation coefficient −0.03 and P-value 0.38) and TCGA melanoma samples (correlation coefficient 0.0564 and P-value 0.24).

One of the limitations of this study is the relative small sample size for some rare tumor types and clinical variable association in this case is less reliable compared to the major tumors where several hundreds to a thousand samples are available. Additionally, some tumor types do not have paired normal samples for comparison such as brain and gastric tumors. As MCAM can be expressed in endothelium and tumor cells at the same time, it was not possible to distinguish the two from RNA-seq data in a bulk tissue. The amount of blood vessels can differ from tumor to tumor. Such heterogeneity could not be accounted for. Lastly, the survival association of MCAM with patient survival was limited to overall survival and univariate analysis as disease-specific survival data was very incomplete and there was a lack of common prognostic factors across different tumor types for adjustment. However, univariate association generally provides good estimate for a potential important marker. To validate this, we selectively evaluated MCAM’s independent prognostic value in lung adeno- and squamous cell carcinoma by adjusting patient age, gender and pathological stage and in both cases no significance was found as reported from univariate analysis.

Some of the results from this study are pretty interesting and provide future directions of study. For example, brain tumors (LGG and glioblastoma multiforme) had consistently increased MCAM expression and were associated with poor outcome of the patients. A potential diagnostic or prognostic marker needs to be further explored in these tumors for clinical applications. Increased expression of MCAM in MESO, a special type of lung cancer, was also significantly associated with poor survival. A study is planned to use archived tissues of MESO for MCAM IHC stain to validate the finding. Results from this study are mostly observational and how MCAM affects tumor behaviors is little known. We plan to conduct laboratory experiments on cell lines and explore its impacts on cell growth, invasions or metastatic abilities by boosting or inhibiting its expression.

In summary, our comprehensive analysis shows that MCAM is indeed highly expressed in some tumors and is associated with tumor progress (tumor stage) and patient survival. However, it is not as widespread as reported, particularly for some common tumors like lung cancer, breast cancer and colon cancer. Although MCAM is consistently highly expressed in SKCM, higher stage tumors do not have a significant increase of its expression and higher expression of this gene is weakly associated with worse survival of patients. As MCAM can be expressed by both cancer cells and stromal blood endothelial cells, its role and clinical implications from these different cells can confound each other and further studies need to separate the two for their roles in cancer biology, interaction and clinical implications. MCAM expression is associated with some therapeutic drugs’ responsiveness or resistance and is highly cancer cells dependent. The high expression of MCAM associated with ‘PLX4720’ and ‘dabrafenib’ response may warrant further investigation.

Acknowledgments

The authors would like to thank GTEx, TCGA and CCLE projects for the data access.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No.81472835), People’s Hospital of Zhengzhou University and Mayo Clinic Center for Individualized Medicine.

Yunxia An is an Attending Physician at Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Henan, China.

Nan Wei is a PhD student in Respiratory Medicine at People’s Hospital of Zhengzhou University, Henan, China.

Xiansong Cheng is a Resident Physician at Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Henan, China.

Ying Li is a Chief Physician and Professor in Department of Respiratory Medicine at Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Henan, China.

Haiyang Liu is a Resident Physician at Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Henan, China.

Jia Wang is a Post Doc Research Fellow at Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Henan, China.

Zhiwei Xu is an Attending Physician at Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Henan, China.

Zhifu Sun is a Consultant and Associate Professor in Department of Health Sciences Research, Mayo Clinic, Rochester, MN, USA.

Xiaoju Zhang is a Professor, Chief Physician and Director of Department of Respiratory Medicine, Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Zhengzhou, Henan, China.

References

Author notes